Bard was trained on a massive dataset of text and code, comprising of 1.56 trillion words of public dialog data and web text. Here’s a breakdown of the data sources that were used to train Bard: The data was carefully…